Rhythm is a circle, and the bar line is a choice
Staff notation draws rhythm as a line running left to right, divided by bar lines into equal groups, with the first beat of each group carrying a special weight.
Every part of that is a decision, and two of them are not obviously right.
A repeating rhythm is a loop. It has a period and a set of positions where something happens, and it goes round. A line has a start and an end; a loop does not, and the decision to cut it at a particular point and call that point “one” is made by a notator, a listener, or a tradition — and different traditions cut it differently.
What rotation does
The clearest demonstration is to take one pattern and start it in different places.
The eight-step pattern with onsets at 1, 4 and 7 is the tresillo — the foundation of Cuban music and, through it, of a very large fraction of the world’s popular music. Rotate it to start at position 4 and the result is a pattern with onsets at 1, 4 and 6, which is a different named rhythm. Rotate again and get another.
Same necklace. Same intervals between attacks, in the same cyclic order. Different names, different traditions, different feel — because the feel depends entirely on which onset is heard as first, and that is not in the pattern.
This is exactly the relationship modes have to scales: one set of positions, several starting points, and a family of distinct-sounding objects that are the same object seen from different places. The parallel is not a metaphor. A scale is a selection of positions from twelve arranged in a cycle, and a rhythm is a selection of positions from eight or twelve or sixteen arranged in a cycle, and the mathematics is identical — the same construction generates both.
Metre is a hierarchy, not a grid
Bar lines are not merely marks every so often. They assert a hierarchy of strength, and the hierarchy is what a listener is actually tracking.
In 4/4, beat one is strongest, beat three is next, beats two and four are weak, and the subdivisions between them are weaker still. That is a tree, and every note in a bar has a position in it.
The tree explains things a flat grid cannot. Why a note tied across a barline sounds like a syncopation: it is sounding at a weak position and being held through a strong one. Why 6/8 and 3/4 are different despite both having six eighth notes: they have different trees, one grouping 3+3 and the other 2+2+2. Why a phrase that ends on beat four feels unfinished.
Additive metres — 3+2, 3+3+2, 2+2+2+3 — do not fit a tree that divides evenly, and traditions using them treat the bar as a sequence of unequal beats rather than as an equal division. Balkan music does this constantly, and a notation designed around even division records it awkwardly.
The downbeat is not in the sound
Here is the strongest claim available, and it is well supported experimentally: where the downbeat is cannot always be recovered from the sound.
Play a listener an unaccented repeating pattern and ask them to tap the beat. Listeners disagree with each other, and individual listeners will hear the same recording with different downbeats on different occasions. Once a downbeat is chosen it is remarkably stable and hard to dislodge — but the choosing was not determined by the acoustics.
Many traditions place the emphasis differently from where a European notation would. In much West African music the organising reference is a repeating bell pattern rather than a metrical downbeat, and several parts enter at different points in the cycle without any of them being “wrong”. The question “where is beat one” may have no answer, and transcriptions into staff notation have to invent one.
What survives rotation
If the starting point is arbitrary, something must be left that is not, or rhythms would be indistinguishable from their rotations. What survives is the interval structure: the sequence of gaps between onsets, read cyclically.
For the tresillo, that sequence is 3–3–2. Every rotation of it — 3–2–3, 2–3–3 — is the same cyclic sequence read from a different place, and it is drawn as the same polygon.
That polygon is a genuinely useful invariant. Two rhythms with the same polygon are rotations of each other and will sound related to any listener; two with different polygons will not, however similar their notation looks. It is the rhythmic equivalent of interval content, and it is invisible in staff notation.
Notation hides the cycle
Staff notation is extremely good at some things and this is not one of them.
The page shows durations in sequence, tied to a metre with a hierarchy that this pattern does not have. A player reading it gets the right result. A reader looking for structure gets a sequence of dissimilar note values in which the underlying evenness is entirely obscured.
Cyclic notations exist and are used where they suit: box notation in percussion pedagogy, the circular diagrams of ethnomusicology, and the grid of every drum machine ever built. That last is worth noticing — the moment somebody had to design an interface for entering rhythms, they produced a grid of equal cells, not a stave.
Counting, and what a pulse is
Underneath every rhythm is a pulse, and a pulse is a stranger thing than it looks.
A pulse is a regular series of moments that a listener projects onto the music. It need not be sounded — a listener taps along to passages where nothing lands on the beat at all — and it survives interruptions, which is how a piece can stop for a bar and start again in time.
The range over which this works is narrow and measurable. Below about 40 events a minute the sense of a connected pulse fails and events are heard as isolated. Above about 600 a minute individual events fuse and the pulse is heard at some subdivision instead. Between those, and most strongly around 100 events a minute, listeners entrain readily and accurately.
That preferred region is suspiciously close to a walking pace and a resting heart rate, and the coincidence has been made much of. What is solidly established is narrower: that the preference exists, that it is consistent across listeners, and that most of the world’s dance music sits in it.
Syncopation is a violation of the tree
With the hierarchy in hand, syncopation stops being a vague term.
A syncopation is an event at a weak position in the metrical tree, made prominent — by accent, by length, by being tied over a strong position that then receives nothing. The effect depends entirely on the listener having the tree in mind: a note off the beat is only off the beat if there is a beat to be off.
That is why syncopation cannot be the norm. A style in which every event is displaced provides nothing to displace from, and the displacement stops being audible. Syncopated music needs a strongly established metre precisely so that it can contradict it, which is why so much of it also contains a completely unsyncopated part holding the metre down.
The tresillo is instructive here. Against a four-beat bar its three onsets fall on beats 1, the second half of 2, and 4 — one on the beat, two off it. It is a syncopation that has become a rhythm in its own right, which is what happens when a displacement is used often enough to become a pattern.
Whose music, and when
Cyclic conceptions of rhythm are the norm rather than the exception, and European linear notation is the outlier.
Indian classical music organises time by tala: a cycle of a fixed number of beats — Teental has sixteen, Jhaptal ten — with an internal structure of stressed and unstressed sections, and a defined point called sam where cycles coincide and to which improvisations return. The cycle is the unit, and a performance is a very large number of turns of it.
West African drumming ensembles are organised around a repeating bell pattern that functions as a temporal reference for parts that enter at different points in it. Arab and Turkish music uses iqa and usul — rhythmic cycles defined by patterns of low and high drum strokes.
The European bar line, by contrast, dates from around 1600, arriving with keyboard tablature and score alignment rather than with any rhythmic idea. It began as a visual alignment aid and acquired its metrical meaning afterwards, which is a good reminder that a notation’s structure and the music’s structure are not the same thing.
Tempo changes what a rhythm is
One parameter is absent from every diagram here and it changes the answer more than any of the ones that are present.
The same pattern at 60 events a minute and at 240 is not the same perceptual object. Slow enough and the listener hears a sequence of separate events with no grouping at all. Fast enough and the whole pattern fuses into a single gesture, and the individual onsets stop being countable.
In between, the level a listener entrains to shifts. Given a pattern in 4/4 at a very slow tempo, listeners tap the subdivisions rather than the beats; at a very fast one they tap the bars. The notation is unchanged and the perceived metre is not — which means the tree in a listener’s head is not the tree on the page.
That has a practical consequence that composers exploit constantly: writing something in 4/4 at a tempo where the bar is the felt beat is a way of getting a slow metre with fast notation, and the reverse is how a fast movement in 3/4 becomes one-in-a-bar.
Where the model stops
Equal steps. Every cycle drawn here divides the period into equal arcs. Real performances do not: swing, groove and microtiming displace onsets by amounts that are musically essential and that a grid has no positions for.
Onsets only. The figures mark where notes start. Duration, dynamics, timbre and articulation all carry rhythmic weight and none of them is drawn.
One cycle length. Music routinely superimposes cycles of different lengths, which is a separate subject and which a single circle cannot show.
No tempo. The same pattern at 60 and at 200 beats a minute is perceptually a different thing — below about 40 events a minute the sense of pulse disappears entirely, and above about 600 individual events fuse. The cycle diagram is scale-free and hearing is not.
The grid is a claim about the music
Every figure on this page places onsets on equally spaced positions, and that is a substantive assumption rather than a drawing convention.
Real performances do not sit on a grid. Measurements of jazz drummers show the second eighth note of a swung pair landing anywhere from 55 to 70 per cent of the way through the beat, varying with tempo, with the player, and within a single performance. Measurements of Viennese waltz playing show the second beat consistently early by a few per cent. Measurements of West African ensembles show systematic offsets between parts that are stable, reproducible and not on any subdivision.
None of that is error. It is reproducible, it is characteristic of players and traditions, and removing it — by quantising a performance to a grid — produces something that musicians describe as dead. A drum machine playing exactly on the grid sounds like a drum machine, and the entire history of programmed rhythm since the 1980s is a history of putting the deviations back.
So the cycle diagram describes the pattern and not the performance. It is the right picture of which positions are struck and the wrong picture of when, and the gap between the two is where a great deal of what people mean by feel lives.
Why the circle is the better diagram anyway
Having granted its limits, the case for the cyclic picture over the linear one is worth making plainly, because it is not merely a matter of taste.
A cycle shows the interval structure, which is what survives rotation and is therefore what two related rhythms have in common. A line shows a sequence of durations, in which the same information is present and unreadable.
A cycle makes rotation visible as a rigid motion — turn the picture and the related rhythm is there. On a line, a rotation is a complete rewrite in which every note value appears to change.
A cycle makes evenness visible as geometric regularity. The maximally even patterns look regular; the irregular ones look irregular. On a line, both look like arbitrary sequences of quavers and crotchets.
And a cycle does not privilege a starting point, which is the honest position for a repeating pattern whose starting point is a listener’s decision.
The linear stave is better at other things — pitch, simultaneity, and telling a player what to do next — which is why it won. It is worth knowing that the choice cost something.
The ladder from here
Later rungs: Euclidean rhythms and the algorithm that generates them. Polyrhythm and the least common multiple. Metre as hierarchy in full. Additive and aksak metres. Swing, and what a straight eighth is not. Microtiming measured. Tala and the cycle in Indian music. The bell pattern as reference. Beat induction, and how listeners find a pulse. Syncopation as a violation of a tree. And the question of whether metre is perceived or imposed, which the tapping experiments make uncomfortably clear.
The drum machine grid — sixteen boxes in a row, click to fill — is the most widely used rhythmic notation in the world by a wide margin, and it is a cycle drawn straight, with the wrap-around implied. Nobody designed it from theory. It is what the problem looks like when the requirement is to enter a repeating pattern quickly.